arctic-annotations-v1

Annotations only — no images, no video.

VITRA-style hand episodes for ARCTIC, with per-hand instructions and paraphrases.

episodes	12,610
training samples (index_frame_pair rows)	425,796
annotation	MANO pose + world/camera joints + per-frame extrinsics
text	one instruction per episode + 1.86 paraphrases on average
source frame rate	30 fps
images / video	not included — see Getting the frames below

ARCTIC is the one dataset here that was built for two-handed manipulation of articulated objects — 11 objects with moving parts, captured while both hands operate them together.

What we did

Episodes are ours. The source release ships either raw video or differently-segmented clips, so we re-cut it with VITRA's method — speed minima of the 3D wrist in world space:

gaussian smooth (sigma=1.0) -> local speed minima in a fixed window (win=15, i.e. 0.5 s
at 30 Hz) -> merge runs shorter than min_seg=16 -> pad 2 frames on each end

sigma and win are quantities in time, converted per source frame rate. Left and right hands are cut independently, with the other hand's motion ignored.

Instructions are ours. Two rounds, both with Qwen3.5-122B-A10B-FP8: round 1 captions 8 frames per episode with the palm's future trajectory drawn on them; round 2 checks the sentence belongs to that hand, strips same-hand references ("Rinse the right hand." -> "Rinse the hand.", because training already prepends Left hand: ... Right hand: ...), and writes 1-3 paraphrases.

Episodes with no instruction are not included. Round 1 returns N/A when an episode shows no object interaction. Those episodes are excluded from both the archive and the index, so every episode here has a usable instruction.

Egocentric view only

ARCTIC records each sequence from 8 static cameras plus one head-mounted egocentric camera. Only the egocentric view is here; we train on head-mounted footage.

The public release covers 301 sequences from 9 subjects (the paper reports 393 from 10; s03 is withheld for the official test server), which is what these episodes are cut from.

One practical note if you fetch the images: ARCTIC's cropped image package does not crop the egocentric view. scripts_data/crop_images.py crops only views 1-8 to 1000x1000 around the object; view 0 is the full frame resized by EGO_IMAGE_SCALE = 0.3, giving 840x600 with the geometry intact. Scale the intrinsics by 0.3 and they match — there is no need for the 649 GB full-resolution package.

Three conventions we converted, and why

Values here follow the same conventions as the other repos in this collection, which differ from ARCTIC's raw MANO parameters in three places. All three were measured, not assumed:

• transl_worldspace is the wrist joint, obtained by running MANO forward and taking joints[0]. ARCTIC's raw trans is not the wrist — it sits 9.23 cm (right) / 9.27 cm (left) away from it.
• hand_pose has hands_mean added back. Round-tripping with it reproduces ARCTIC's own joints to 0.000000 cm.
• The left hand uses a genuine MANO_LEFT model, not a mirrored right hand.

Also checked: 21-joint bone-length CV is 0.00000 (joint order matches ours), median bone length 27.8 mm (TACO 28.2, EPIC 29.0), extrinsics deviate from orthonormal by 8.1e-07, and MANO frame counts equal egocentric frame counts on every sequence tested.

Files

arctic.tar                  ->  Annotation/arctic/episodic_annotations/*.npy
episode_frame_index.npz        index_frame_pair (N,2) uint32 + index_to_episode_id (E,)

index_frame_pair row number is the sample id: row r = (episode ordinal, frame within that episode). len(index_frame_pair) is the size of the training set.

import numpy as np
# tar -xf arctic.tar
z = np.load("episode_frame_index.npz", allow_pickle=True)
ep_slot, frame_id = z["index_frame_pair"][sample_id]
eid = str(z["index_to_episode_id"][ep_slot])
d = np.load(f"Annotation/arctic/episodic_annotations/{eid}.npy", allow_pickle=True).item()
rgb_frame_id = int(d["video_decode_frame"][frame_id])

Each .npy is a dict with video_name, video_decode_frame, intrinsics, per-frame extrinsics (world->camera), anno_type (which hand this episode is for), text, text_rephrase, and a left/right dict holding beta, hand_pose, global_orient_worldspace, transl_worldspace, joints_worldspace, kept_frames. text[hand] = [(sentence, (0, T))] and text_rephrase[hand] = [([paraphrases...], (0, T))].

Getting the frames

video_decode_frame indexes the source video, which we do not redistribute. Get it from ARCTIC — https://arctic.is.tue.mpg.de/, then decode by index (we use decord; a self-maintained sequential counter drifts silently if the decoder ever skips a frame).

Known limitations

• Paraphrase count averages 1.86, not a fixed number. Past 3 the model starts inventing; a sentence with no prepositional phrase honestly supports only one or two.
• ARCTIC is a lab capture: a mocap studio, 11 objects, 9 subjects. Its pose quality is the highest here, its visual and task diversity the lowest. It is 2.3 hours of source video.
• Verified: the index lists exactly the episodes that have an instruction, every episode's stored frame count matches its index rows, and no index entry points at a missing episode.

---

The collection

Every dataset we have taken through this pipeline, with what is published today. All repos live under MIT-Media-Lab and are annotations only — no images, no video.

dataset	episodes	training samples	our contribution	size	HF
EPIC-KITCHENS-100	149,570	4,019,534	episodes + text	8.70 GB	epic30-annotations-v1
EgoTouch	111,159	3,687,389	episodes + text + tactile	20.12 GB	egotouch-annotations-v1
GigaHands	70,486	2,266,087	episodes + text	2.89 GB	gigahands-annotations-v1
Ego-Exo4D	67,051	1,757,474	text only	4.09 GB	egoexo4d-annotations-v1
Something-Something V2	52,706	1,124,722	text only	4.63 GB	ssv2-annotations-v1
OakInk2	28,264	1,371,721	episodes + text	1.92 GB	oakink2-annotations-v1
TACO	23,757	736,136	episodes + text	1.34 GB	taco-annotations-v1
ARCTIC	12,610	425,796	episodes + text	0.85 GB	arctic-annotations-v1
H2O	5,696	196,941	episodes + text	0.40 GB	h2o-annotations-v1
total	521,299	15,585,800		44.9 GB

episodes = entries in episode_frame_index.npz, i.e. what a training run actually sees. training samples = rows of index_frame_pair; the row number is the sample id.

episodes + text means we re-cut the source ourselves at wrist-speed minima and then wrote the instructions. text only means the episodes are VITRA-1M's official segmentation, used unchanged, and only the instructions are ours.

Episodes whose round-1 caption came back N/A (no object interaction) are not published — they are excluded from both the archive and the index, so every episode here has a usable instruction. That is why the published counts are below the totals we cut:

	episodes on disk	published	dropped as N/A
EPIC-KITCHENS-100	151,502	149,570	1,932 (1.3%)
EgoTouch	147,386	111,159	36,227 (24.6%)
GigaHands	92,365	70,486	21,879 (23.7%)
Ego-Exo4D	67,051	67,051	0
Something-Something V2	52,706	52,706	0
OakInk2	37,692	28,264	9,427 (25.0%)
TACO	26,454	23,757	2,697 (10.2%)
ARCTIC	14,821	12,610	2,211 (14.9%)
H2O	7,792	5,696	2,096 (26.9%)

ssv2 and egoexo4d are 0 because VITRA-1M already dropped N/A upstream — their episodes are the official segmentation, so there was nothing left for us to drop. Their on-disk counts are slightly below VITRA-1M's published index (52,718 and 67,053) because round 2 marked a handful of sentences unusable and we deleted those episodes: 12 from ssv2, 2 from egoexo4d.

DexYCB was removed

DexYCB was removed from this collection on 2026-08-30. It is captured by 8 fixed RealSense cameras around a table; its own camera.role field reads allocentric on all 15,878 episodes. Unlike OakInk2, which ships an egocentric view alongside three allocentric ones, DexYCB has no head-mounted camera at all, so there was nothing to filter down to.

Not published yet

dataset	episodes cut	where it stands
HOI4D	—	source converted by a colleague; not re-cut
HOT3D	—	source converted by a colleague; not re-cut